PrismML unveils compact Bonsai LLM model for smart glasses on the Snapdragon AR1 Gen 1 chip
PrismML has introduced the 1-bit Bonsai LLM with 2 billion parameters for smart glasses on Qualcomm's Snapdragon AR1 Gen 1 platform — the model runs locally without sending data, though no specific glasses using it have been announced yet.
PrismML, a company founded by researchers from Caltech with Ion Stoica of UC Berkeley serving as an advisor, has unveiled a version of its small language model, the 1-bit Bonsai LLM, for smart glasses built on Qualcomm's Snapdragon AR1 Gen 1 platform. The model was shown on Wednesday at the Qualcomm Snapdragon Summit.
According to PrismML, the model shrinks larger models by roughly four times, while, according to its claims, retaining nearly all performance on standard benchmarks. The smart glasses version has 2 billion parameters and is tuned for simultaneous vision and language processing, allowing the user to ask in real time about what they are currently seeing. The model runs locally directly on the chip, with no need to send data outside the device.
PrismML's stated goal is to offer open-weight AI running on devices that makes better use of the computing power these devices already have, as an alternative to dependence on proprietary cloud AI labs and their growing need for computing power. The release of the model for Qualcomm's chip is a step in this direction, though no specific smart glasses using this software have been announced yet.
Why it matters
It illustrates a trend toward AI models for wearable devices running directly on the chip instead of in the cloud, which addresses privacy and latency concerns. For hardware makers and AR app developers, this represents a documented technical choice versus cloud-based solutions, albeit without a specific commercial product yet.
Relevant practical impact
What this means
For a business
For companies developing wearable hardware or AR applications, a documented alternative to dependence on cloud AI services is emerging: a model running locally on the chip without the need to send data elsewhere, which could influence decisions on product architecture as well as cloud inference costs.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.